Revisiting Non-Verbatim Memorization in Large Language Models: The Role of Entity Surface Forms
For researchers evaluating LLM factual knowledge, this work highlights that current entity-based QA benchmarks may overestimate memorization by using only canonical names, and provides a resource to assess surface-form robustness.
The paper introduces RedirectQA, a dataset linking Wikidata facts to multiple surface forms of entities, and shows that LLM factual recall varies significantly with entity name variations, with robustness depending on variation type and frequency.
Understanding what kinds of factual knowledge large language models (LLMs) memorize is essential for evaluating their reliability and limitations. Entity-based QA is a common framework for analyzing non-verbatim memorization, but typical evaluations query each entity using a single canonical surface form, making it difficult to disentangle fact memorization from access through a particular name. We introduce RedirectQA, an entity-based QA dataset that uses Wikipedia redirect information to associate Wikidata factual triples with categorized surface forms for each entity, including alternative names, abbreviations, spelling variants, and common erroneous forms. Across 13 LLMs, we examine surface-conditioned factual memorization and find that prediction outcomes often change when only the entity surface form changes. This inconsistency is category-dependent: models are more robust to minor orthographic variations than to larger lexical variations such as aliases and abbreviations. Frequency analyses further suggest that both entity- and surface-level frequencies are associated with accuracy, and that entity frequency often contributes beyond surface frequency. Overall, factual memorization appears neither purely surface-specific nor fully surface-invariant, highlighting the importance of surface-form diversity in evaluating non-verbatim memorization.